Lightweight Low-Light Image Enhancement Techniques for Edge Devices

The field of computer vision has a lot of practical applications in today's world, such as surveillance, autonomous navigation, medical imaging, etc. However, these computer vision systems frequently encounter issues such as images captured under poor lighting conditions, low contrast, backlit images, and noise infiltration into the image. The enhancement of these dark and dimly lit images is crucial for maintaining the reliability and accuracy of computer vision systems. This paper provides a comprehensive overview of traditional and deep learning-based methods for low-light image enhancement. We thoroughly explore traditional methods such as grey level transformation, gamma correction, histogram equalization, and Retinex theory-based methods, highlighting their effectiveness and limitations. In addition, the paper discusses the suitability of the methods for edge deployment. The paper also proposes a model for the detection and classification of low-light images on edge devices.